CN104408456A - Hough transformation linear detection method based on dynamic threshold range - Google Patents
Hough transformation linear detection method based on dynamic threshold range Download PDFInfo
- Publication number
- CN104408456A CN104408456A CN201410591811.3A CN201410591811A CN104408456A CN 104408456 A CN104408456 A CN 104408456A CN 201410591811 A CN201410591811 A CN 201410591811A CN 104408456 A CN104408456 A CN 104408456A
- Authority
- CN
- China
- Prior art keywords
- interval
- pixel
- value
- point
- line
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/48—Extraction of image or video features by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a Hough transformation linear detection method based on a dynamic threshold range. The method comprises the following steps: performing storage in accordance with a cluster sequence based on adjacent relation, i.e., performing refinement processing on segments with certain line widths in images, outputting a frame line which has only two adjacent pixel points at most in eight neighborhoods of any one pixel point, and then based on the cluster sequence based on the adjacent relation, storing the frame line in a certain segment of arrays; and sampling m adjacent pixels in the certain segment of arrays to construct the threshold range by sampling, determining whether m points are at the same line, and filtering noise at the same time, such that the algorithmic robustness is enhanced. On the basis of constructing the threshold range, linear detection is realized by use of a dynamic sampling mode based on range halving. Each pixel in the pixel arrays is identified, such that the algorithmic detection efficiency and effect are improved.
Description
Technical field
The invention belongs to image characteristics extraction and recognition technology field, particularly a kind of Hough transform line detection method based on dynamic threshold interval.
Background technology
Image characteristics extraction is the prerequisite of figure identification, relates to the multi-door ambits such as computer graphics, Digital Image Processing, pattern-recognition, artificial intelligence, has wide research space and application prospect.Engineering drawing Intelligent Recognition is the important applied field of image characteristics extraction.In all kinds of engineering drawing, the main geometric element of engineering pattern object is line segment.How fast and effectively Checking line feature is the basis identifying Drawing Object.
Lines detection method is mainly divided into two large classes: first kind method is based on geometric properties extracting directly straight line in image area, utilizes the information such as the gradient of pixel in image space or gray scale to process.Equations of The Second Kind method is transformed in parameter field by image to extract straight line, indirectly process according to the information in parameter field, if Hough transform is in Hough parameter field by the reflection of graphics, according to a certain analytic expression each spot projection on figure in parameter field, and carry out simple cumulative statistics, then find cumulative peakvalue's checking straight line.Although based on the method speed of image area, be easy to the interference being subject to noise, there is discontinuous straight line, limit practical application.Method noise immunity based on parameter field is high, and can connect conllinear short lines, is suitable for parallel processing.But the method carries out image area to map to the one-to-many of parameter field, and there is the impact of noise spot and discretization error, cause that calculated amount is large, peak point is surrounded by minor peaks point and causes undetected or flase drop in parameter field, threshold value is difficult to the problems such as setting.
Large for Hough transform calculated amount, A.Goldenshluger and A.Zeevi adopts random Hough transformation method, randomly draw a pixel in a plurality of pixels and be mapped to parameter space as sample information, avoid compared with intensive and more memory cost, but because the method obtains sampled point at random, thus cause a large amount of invalid accumulations, reduce the efficiency of method.
For the problem that peak point in parameter field is surrounded by minor peaks, the people such as Siyu Guo propose the method that neighborhood suppresses, be weighted projecting to ballot value in parameter field, the impact of effective reduction minor peaks, improve the accuracy rate that Hough transform detects, but the setting of the weighted value of the method has a significant impact to testing result, needs a large amount of test just can obtain the exact value of weight, add the computation complexity of the method, reduce the practicality of the method.
Summary of the invention
In order to overcome the deficiency of method above, the present invention proposes a kind of Hough transform line detection method based on dynamic threshold interval.Step is as follows:
Step 1: the cluster sequential storage based on syntople: carry out thinning processing for the line segment in image with certain live width, the skeleton line of two neighbor pixels is only had at most in 8 neighborhoods exporting arbitrary pixel, then carry out the cluster sequential storage based on syntople, the pixel on skeleton line is stored in a certain hop count group.
Step 2: threshold interval builds: m the adjacent pixels of sampling in a certain hop count group, builds threshold interval, judges that m is put whether on same straight line, carries out the filtration of noise simultaneously, enhances the robustness of method.
Step 3: dynamically straight-line detection: for a certain section of array of pixels, on the basis building threshold interval, adopts the dynamic sampling mode based on interval halving to realize the detection of straight line.
Step 4: straight line splices, according to the slope of this straight-line segment and the shortest adjacent distance size of end points, the adjacent short lines section after detecting for said method, judges whether two line segments belong to a line segment, realizes the optimal classification of straight-line segment like this.
Step 5: terminate.
The present invention compared with prior art has the following advantages:
(1) straight line in image is subject to discretize and noise effect, to cause in parameter threshold space main peak value by minor peaks surround cause undetected with flase drop etc. problem, propose the concept in dynamic threshold interval, avoid the generation of above-mentioned situation, the robustness of increase method.
(2) after structure threshold interval, only need calculating just can judge whether on straight line for twice to detected pixel, avoid a large amount of calculating that traditional Hough detects.
(3) in the straight-line detection process of a certain section of array of pixels, adopt the dynamic sampling mode detection threshold based on interval halving interval, and each pixel in array of pixels is identified, improve detection efficiency and the effect of method.
Accompanying drawing explanation
Fig. 1 is a kind of Hough transform line detection method straight line cluster sequential storage process flow diagram based on dynamic threshold interval of the present invention;
Fig. 2 is the Hough transform straight line parameter threshold interval design of graphics of a kind of Hough transform line detection method m sampled point based on dynamic threshold interval of the present invention;
Fig. 3 is the Hough transform straight-line detection schematic diagram of a kind of Hough transform line detection method dynamic sampling based on dynamic threshold interval of the present invention;
Embodiment
Also in conjunction with specific embodiments the present invention is further elaborated referring to accompanying drawing of the present invention, but protection scope of the present invention not limited by concrete embodiment, is as the criterion with claims.In addition, with without prejudice under the prerequisite of the present invention program, any change that those of ordinary skill in the art made for the present invention easily realize or change all will fall within right of the present invention.
The present invention includes following committed step:
Step 1: after the refinement of image cathetus, only has at most the skeleton line of two neighbor pixels in 8 neighborhoods exporting arbitrary pixel, adopt the flow process shown in accompanying drawing 1 by straight-line segment sequential storage in a certain hop count group;
Step 2: build threshold interval;
(1) structure in noise-free case lower threshold value interval
Under normal conditions, for be in straight line inclination angle be α (such as, straight line inclination alpha=60.5 ° of Fig. 2 (b)) the individual adjacent ideal point of m (m=9), the interval ρ of ρ value (θ) is as shown in Fig. 2 (a).Due to the interference of sampling angle θ quantization error, the result being mapped to parameter field drops on left and right adjacent sample values θ
1, θ
2between: θ
1=[α]+90 °, θ
2=[α]+90 °+1 ° ([] is for rounding).
Due to α=60.5 °, then θ
1=150, ρ
1=[ρ
l1ρ
u1] and θ
2=151, ρ
2=[ρ
l2ρ
u2], ρ
1m noiseless sampled point θ in parameter field
1when=150, the set of corresponding ρ value composition is interval, in like manner ρ
2θ
2when=151, the set of corresponding ρ value composition is interval, as shown in Fig. 2 (c), is proved the ρ in Hough transform by theoretical and experiment
1with ρ
2meet following three constraints:
1. ρ
1with ρ
2m point the shortest and secondary two short intervals of length in sampling angle ρ (θ) (as Fig. 2 (a)) in parameter field;
2. ρ
u2< ρ
l1, and be m point on α straight line for being in inclination angle, sampling angle θ in Hough parameter field
αthe interval of upper correspondence is necessarily included in [ρ
u2ρ
l1] in interval;
3. inclination angle be α straight line on after any point is mapped to parameter field, θ
αadjacent ρ interval [ρ after corresponding discretize
θ 1ρ
θ 2] certain and [ρ
u2ρ
l1] interval crossing.
(2) structure in noise situations lower threshold value interval is had
Consider that the noise in image also has interference to straight-line detection, Fig. 2 (d) shows the noise in image point (D in Fig. 2 (b)
1, D
2) interference profile in parameter field, away from the sinusoidal dotted line P that below normal ρ value is interval in Fig. 2 (d)
1, P
2with noise spot, there is corresponding relation.Due to P
1, P
2in
with
interference, makes ρ
1with ρ
2no longer meet 1., 2., characteristic 3..Therefore, in this paper noise reduction process process, need curve P corresponding for noise spot
1, P
2filter out, and determine θ
1, θ
2.Here by θ
1, θ
2, ρ
1with ρ
2determined interval is as the straight line parameter information of m sampled point.
The filtration of noise information and the assignment procedure in parameter threshold interval:
1. obtain m continuous image vegetarian refreshments from sampling location, in parameter field, calculate the sinusoidal curve that m point is corresponding;
2. for all sampling angle θ, calculate ρ set interval, find the shortest and interval ρ of secondary short set
1with ρ
2;
3. whether overlapping both judging, if the two is not overlapping, then determine [ρ
u2ρ
l1], θ
1with θ
2as parameter threshold.Otherwise turn 4.;
4. process ρ
1, ρ
2the P on border is in set
1, P
2deng the sinusoidal dotted line that noise spot is corresponding, remove, record the number n removing sinusoidal dotted line, this step is described in detail as follows simultaneously:
ρ respectively
1, ρ
2depart from the frontier point of normal ρ value in set, the two is ρ simultaneously
1, ρ
2coboundary or lower boundary (as Fig. 2 (d) is depicted as coboundary).The computation process finding frontier point is as follows: first calculate ρ
1, ρ
2the center of gravity of set, and then calculate the distance of up-and-down boundary point and center of gravity, being worth large person is
noise spot, thus determine corresponding sinusoidal dotted line P
1, after removing this dotted line (n is the count value after m removes noise spot number), rejudge ρ
1, ρ
2whether set interval is overlapping, if not overlapping, turns back to 2.; If overlapping, first judge whether n is greater than [m/2], if very, exit parameter threshold assignment procedure, think that m sampled point is not point-blank; Otherwise repeat this step;
Step 3: based on the straight-line detection of dynamic sampling;
Identify in establishing method that the array index of the Origin And Destination of a certain section of pixel in array is S, E (corresponding skeleton line is as Suo Shi Fig. 3 (a)).
1. its initial value is that (rectangular pixels point number) – 1, as shown in Fig. 3 (b) for S=0, E=length.
2. adopt detection of dynamic principle, judge array index centre position H
0whether=[(S+E)/2.0] place m neighbor is on same straight line, and determination methods describes in the parameter threshold interval of step 2 builds.If m pixel is not point-blank, then with H
0point is cut-point, is divided into two sections of pixel intervals: array index is from S to H
0, H
0to E, to every section of interval pixel recursive call this method.If H
0on position, m point is on same straight line L, then whether on the linel to judge each pixel left and to the right respectively, 3. deterministic process turns.
3. for each pixel (x, y), basis for estimation be constraint condition in step 2 3..In the process judged left, if 3. this pixel satisfies condition, then think on straight line, straight line classification information is stored in the 2nd dimension of array; And continue in turn to judge next pixel.If not on straight line, be then interval recursive call this method that the point of S forms to this point to array index, turn 1..
Whether, in said method processing procedure, record each pixel and be processed, if process, value is 1, otherwise value is zero.From S to H
0after interval pixel has judged, then Recursion process H
0to the interval of E.The method judge end condition be complete to all processes pixel in array after, forward next step to and carry out straight line merging.
In accompanying drawing 3 (c), H
i(i=0,1,2) are the subscript positions of m pixel of sampling successively in method, L (H
i) (i=0,1,2) be the straight line identified.
Step 4: whether whether be less than 3 pixels with end-point distances according to straight-line segment inclination angle difference within ± 1 ° and judge whether two straight lines can merge into straight line.The straight line identified in fig. 3 does not meet above-mentioned condition, cannot merge.
Step 5: terminate.
Claims (7)
1., based on the Hough transform line detection method in dynamic threshold interval, it is characterized in that comprising following step:
(1) based on the cluster sequential storage of syntople;
(2) threshold interval based on Hough transform builds;
(3) adopt the straight-line detection of dynamic sampling, for a certain section of array of pixels, carry out dynamic straight-line detection;
(4) straight line splicing;
(5) terminate.
2. detection method as claimed in claim 1, is characterized in that, in step (1):
Export the skeleton line of single pixel live width after image thinning, on skeleton line arbitrary pixel 8 neighborhoods in only have at most two neighbor pixels;
This algorithm carries out cluster according to pixel syntople, by the class picture element position information (line position after cluster
x, column position
y) be sequentially stored in one section of array of pixels.
3. detection method as claimed in claim 1, it is characterized in that, step (2) builds based on the threshold interval of Hough transform, is divided into the following steps:
for in one section of array of pixels
mindividual neighbor, right in Hough transform domain according to formula
θget 0,1 ..., the sampled values such as 179, calculate sinusoidal sampled point (
θ, ρ);
for
m(initial value
m=
m) bar sinusoidal curve, one
θvalue is by correspondence
mindividual
ρvalue, calculates
mindividual
ρvalue range intervals [
ρ l ,
ρ u ] (
ρ l for floor value,
ρ u for upper dividing value), setting should
ρvalue burst length is
l ρ ;
for all sampling angles
θ, calculate respectively
l ρ , and find minimum length
l ρ1
's
ρ 1value interval [
ρ l1
,
ρ u1
], record corresponding angle is
θ 1; To calculate adjacent angular again
ρvalue is interval, finds time short length
l ρ2
's
ρ 2value interval [
ρ l2
,
ρ u2
];
judgement interval [
ρ l1
,
ρ u1
] with [
ρ l2
,
ρ u2
] whether overlapping, if not overlapping, then the interval section had therebetween is configured to threshold interval [
ρ l_
θ ,
ρ u_
θ ], threshold interval builds and terminates; If the two is overlapping, be then divided into
mexist in individual some noise spot or
mindividual point not on same straight line two situations consider, turn
process;
first consider
mthere is the situation of noise spot in individual point, calculate
ρ 1with
ρ 2frontier point in interval and
ρ 1with
ρ 2the distance of interval center of gravity, large value person corresponds to noise spot, is removed by sinusoidal curve corresponding to this point;
Arrange
m=
m-1, if
m< [
m/ 2] then think
mindividual point, not on same straight line, is chosen again
mindividual point carries out threshold interval structure; Otherwise, turn
circular treatment.
4. detection method as claimed in claim 1, is characterized in that, in step (3), adopt the straight-line detection of dynamic sampling, for a certain section of array of pixels, the step of carrying out dynamic straight-line detection is as follows:
be designated as under supposing to identify in algorithm the Origin And Destination of a certain section of array of pixels
s,
e, its initial value is
s=0,
e=length (array pixel number) – 1;
adopt detection of dynamic principle, judge array index centre position
h 0=[(
s+
e)/2.0] place
mindividual neighbor whether on same straight line, the threshold calculations of determination methods (2) step
middle description;
If
mindividual pixel not point-blank, then with
h 0point for cut-point, is divided into two sections of pixel intervals: array index from
sarrive
h 0,
h 0arrive
e, to every section of interval this algorithm of pixel recursive call;
if
h 0on position
mindividual point is at same straight line
lon, then judge that whether each pixel is at straight line respectively left and to the right
lon, deterministic process turns
;
for each pixel (
x,
y), when judging whether to belong to straight line, only need to calculate
θ 1with
θ 2corresponding
ρ θ1
value interval with
ρ θ2
value is interval, and judges that whether this interval is overlapping with interval, if overlapping, then thinks on straight line, straight line classification information is stored in the 2nd dimension of array; And continue in turn to judge next pixel;
If not on straight line, to array index be then
sinterval this algorithm of recursive call of forming to this point of point, turn
;
Whether algorithm stops Rule of judgment: in algorithm process process, record each pixel and be processed, if process, value is 1, otherwise value is zero;
From
sarrive
h 0after interval pixel has judged, then Recursion process
h 0arrive
einterval, this algorithm judge end condition be complete to all processes pixel in array after, exit algorithm.
5. method according to claim 1, carries out order aggregating storing to the straight line of pixel live width single after image thinning.
6. method according to claim 1, from the pixel set detected, dynamic sampling
mindividual pixel builds threshold interval, and carries out the process of noise filtering.
7. method according to claim 1, for a certain section of array of pixels, carries out dynamic straight-line detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410591811.3A CN104408456A (en) | 2014-10-28 | 2014-10-28 | Hough transformation linear detection method based on dynamic threshold range |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410591811.3A CN104408456A (en) | 2014-10-28 | 2014-10-28 | Hough transformation linear detection method based on dynamic threshold range |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104408456A true CN104408456A (en) | 2015-03-11 |
Family
ID=52646087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410591811.3A Pending CN104408456A (en) | 2014-10-28 | 2014-10-28 | Hough transformation linear detection method based on dynamic threshold range |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104408456A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106371082A (en) * | 2016-08-26 | 2017-02-01 | 上海无线电设备研究所 | Linear velocity pull-off jamming identification method |
| CN109190452A (en) * | 2018-07-09 | 2019-01-11 | 北京农业智能装备技术研究中心 | Crop row recognition methods and device |
| CN113705576A (en) * | 2021-11-01 | 2021-11-26 | 江西中业智能科技有限公司 | Text recognition method and device, readable storage medium and equipment |
| CN116030485A (en) * | 2023-02-20 | 2023-04-28 | 深圳市大乐装建筑科技有限公司 | Method and system for quickly combining collinear line segments based on assembled component detailed diagram |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8081798B2 (en) * | 2007-11-20 | 2011-12-20 | Lawrence Livermore National Security, Llc | Method and system for detecting polygon boundaries of structures in images as particle tracks through fields of corners and pixel gradients |
| CN103871052A (en) * | 2014-02-21 | 2014-06-18 | 杭州奥视图像技术有限公司 | Straight line detection-based safety belt detecting algorithm |
| CN103927541A (en) * | 2014-04-21 | 2014-07-16 | 西北工业大学 | Target edge line detection method applicable to space tethered robot capture process |
-
2014
- 2014-10-28 CN CN201410591811.3A patent/CN104408456A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8081798B2 (en) * | 2007-11-20 | 2011-12-20 | Lawrence Livermore National Security, Llc | Method and system for detecting polygon boundaries of structures in images as particle tracks through fields of corners and pixel gradients |
| CN103871052A (en) * | 2014-02-21 | 2014-06-18 | 杭州奥视图像技术有限公司 | Straight line detection-based safety belt detecting algorithm |
| CN103927541A (en) * | 2014-04-21 | 2014-07-16 | 西北工业大学 | Target edge line detection method applicable to space tethered robot capture process |
Non-Patent Citations (2)
| Title |
|---|
| 宋晓宇等: "基于自适应阈值区间的动态采样Hough变换直线检测算法", 《沈阳建筑大学学报(自然科学版)》 * |
| 宋晓宇等: "基于自适应阈值区间的广义 Hough变换图形识别算法", 《仪器仪表学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106371082A (en) * | 2016-08-26 | 2017-02-01 | 上海无线电设备研究所 | Linear velocity pull-off jamming identification method |
| CN106371082B (en) * | 2016-08-26 | 2019-01-08 | 上海无线电设备研究所 | Linear speed pull-off jamming recognition methods |
| CN109190452A (en) * | 2018-07-09 | 2019-01-11 | 北京农业智能装备技术研究中心 | Crop row recognition methods and device |
| CN113705576A (en) * | 2021-11-01 | 2021-11-26 | 江西中业智能科技有限公司 | Text recognition method and device, readable storage medium and equipment |
| CN116030485A (en) * | 2023-02-20 | 2023-04-28 | 深圳市大乐装建筑科技有限公司 | Method and system for quickly combining collinear line segments based on assembled component detailed diagram |
| CN116030485B (en) * | 2023-02-20 | 2023-07-18 | 深圳市大乐装建筑科技有限公司 | Method and system for quickly combining collinear line segments based on assembled component detailed diagram |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103826102B (en) | A kind of recognition methods of moving target, device | |
| CN102628814B (en) | Automatic detection method of steel rail light band abnormity based on digital image processing | |
| CN103714363B (en) | A kind of motor vehicle exhaust smoke video identification system | |
| JP4832227B2 (en) | Object detection apparatus and object detection method | |
| US9373043B2 (en) | Method and apparatus for detecting road partition | |
| CN104408456A (en) | Hough transformation linear detection method based on dynamic threshold range | |
| CN106650913A (en) | Deep convolution neural network-based traffic flow density estimation method | |
| CN105426864A (en) | Multiple lane line detecting method based on isometric peripheral point matching | |
| CN103164711A (en) | Regional people stream density estimation method based on pixels and support vector machine (SVM) | |
| CN107833206B (en) | Accurate power line extraction method under complex background | |
| CN103605967A (en) | Subway fare evasion prevention system and working method thereof based on image recognition | |
| WO2013053159A1 (en) | Method and device for tracking vehicle | |
| CN102043958A (en) | High-definition remote sensing image multi-class target detection and identification method | |
| CN104156731A (en) | License plate recognition system based on artificial neural network and method | |
| CN105336169A (en) | Method and system for judging traffic jams based on videos | |
| CN104392239A (en) | License plate identification method and system | |
| US11373417B2 (en) | Section line recognition device | |
| CN102156884A (en) | Straight segment detecting and extracting method | |
| Zhai et al. | Crosswalk detection based on MSER and ERANSAC | |
| SE1550006A1 (en) | Method and system for categorization of a scene | |
| CN102592128A (en) | Method and device for detecting and processing dynamic image and display terminal | |
| Li et al. | RailNet: An information aggregation network for rail track segmentation | |
| Xu et al. | Road lane modeling based on RANSAC algorithm and hyperbolic model | |
| CN115690062A (en) | Rail surface damage state detection method and device and electronic equipment | |
| CN110223505A (en) | A kind of determination method and device of pedestrian running red light behavior |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150311 |